Strang, G. #1993#. Wavelet transforms versus Fourier transforms. Bulletin #New Series# of the American Mathematical Society 28, 288-305.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....wavelet frames and orthonormal wavelets. Wavelet frame theory gives an over complete or redundant transform. Orthonormal wavelet theory (also called multiresolution analysis theory) gives an orthonormal, non redundant transform. There are now a number of texts and tutorials available on wavelets [20,31,72,90,91]. For the rest of this chapter, I will concentrate on the discrete orthonormal wavelet transform. 3.1 Discrete Orthonormal Wavelets One Dimension The discrete orthonormal wavelet transform (DWT) is defined as a series of coefficients DWT[m;n] resulting from the projection of a signal f(x) 2 ....

Gilbert Strang. Wavelet transforms versus Fourier transforms. Bulletin of the American Mathematical Society, 28(2):288--305, 1993.

....in Figure 1, in which we use the symbol x to denote the sequence x k and x to denote the time reversed sequence x Gammak . 3. Finally, we note that the DBWT, like its orthogonal counterpart, is an O(n) procedure. This is evident from a factorization of the matrix W n into blocks as described in [2]: W n = 2 6 6 6 4 odd Gamma even shuff le 3 7 7 7 5 1 p 2 2 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 4 a 0 a 1 Delta Delta Delta aN Gamma1 0 0 Delta Delta Delta 0 0 b 0 b 1 Delta Delta Delta b N Gamma1 0 0 Delta Delta Delta 0 0 0 0 a 0 a 1 Delta ....

G. Strang, `Wavelet Transforms Versus Fourier Transforms', Department of Mathematics, Massachusetts Institute of Technology (1992).

....in which we use the symbol x to denote the sequence x k and x to denote the time reversed sequence x Gammak . Finally, we note that the FBWT, like its orthogonal counterpart, can be performed in O(n) operations. This is evident from a factorization of the matrix W n into blocks as described in [3]. 3 Adaptation of Wavelets to Particular Differential Operators In this section we describe the Dahlke Weinreich construction and we make some important observations regarding the properties of the resulting biorthogonal wavelets. 3.1 The Dahlke Weinreich Construction The motivation behind the ....

G. Strang, `Wavelet Transforms Versus Fourier Transforms', Department of Mathematics, Massachusetts Institute of Technology (1992).

....this biorthogonal case in more detail in Section 6.4. The idea of constructing maximally smooth wavelets when some side conditions are specified has been central to much of the activity in wavelet analysis and its applications since the mid 1980 s. In addition to [Dau92] the survey article [Stra93] is enjoyable reading as a backdrop to our book. See also the latter half of the tutorial to Chapter 3. The paper [LaHe96] treats the issue in a more specialized setting and is focussed on the moment method. Some of the early applications to data compression and image coding are done very nicely ....

G. Strang, Wavelet transforms versus Fourier transforms, Bull. Amer. Math. Soc. (N.S.) 28 (1993), 288--305.

....possible combination of filtercoefficients but in fact, it seems that it does. 5. Our algorithm is of complexity O(N) compared to O(N log(N) in the wavelet packet case (see below) Since the computational complexity of the classical wavelet transform is a O(N) operation having N samples ([19]) our algorithm as well has complexity O(N ) at which the constant is of course bigger as in the wavelet case and depends on the size of the library, the number of decomposition levels and wether NSMRA or INSMRA is used. The computation of the cost functions of the detail spaces is an additional ....

G. Strang. Wavelet transforms versus fourier transforms. Bull. Amer. Math. Soc., 28(2):288--305, 1993.

....k x l 2 ) dx = 0 ; 2:1) where ffi ij denotes the Kronecker delta. Assume that the wavelets are of order r, i.e. for some r 1 and 6= 0, we have R OE = 1, R = 0, Z x k OE(x) dx = Z x k (x) dx = 0 for 1 k r Gamma 1 ; Z x r (x) dx = r ) Gamma1 ; see, for example, Strang (1989, 1993). Any square integrable function f admits a generalized Fourier expansion in terms of wavelets: f = X l b l OE l 1 X k=0 X l b kl kl ; 2:2) where, for p 0 and p k = p 2 k , we define OE l (x) p 1=2 OE(px l) kl = p 1=2 k (p k x l) b l = Z f OE l ; b kl = Z ....

.... expansion in terms of wavelets: f = X l b l OE l 1 X k=0 X l b kl kl ; 2:2) where, for p 0 and p k = p 2 k , we define OE l (x) p 1=2 OE(px l) kl = p 1=2 k (p k x l) b l = Z f OE l ; b kl = Z f kl : 5 Analytical details are summarized concisely by Strang (1989, 1993) and Daubechies (1992) The generalized Fourier series in (2.2) converges in mean square. When f is a probability density, and data from the associated distribution are available, an estimator of f may be constructed by developing an empirical version of the expansion (2.2) For example, if fX 1 ....

STRANG, G. (1993). Wavelet transforms versus fourier transforms. Bulletin (N.S.) Amer. Math. Soc. 28, 288--305.

....Also, 5] does not consider biorthogonal wavelets for discrete time periodic signals as we do in Section 5. The connection between wavelets and filter banks has been well documented by Vetterli and Herley [26] Recent expository treatments of wavelets include [6] 10] 12] 13] 20] 23] and [24]. The paper is organized as follows. In the remainder of this section, we introduce discrete time periodic wavelets and compare them to discrete time sinusoids. In Section 2 we review the properties of abstract pyramid type algorithms [4] which are the key to fast wavelet transforms. Section 3 ....

....be more efficient to do this with fast Fourier transforms. For small p, Rioul and Duhamel suggest the use of short length fast running FIR algorithms [16] 25] The fast wavelet transform can also be expressed as a matrix factorization into sparse matrices analogous to the fast Fourier transform [24]. 3.3. Numerical example To obtain the g(n) mentioned at the beginning of the previous subsection, it is necessary to solve (36) 37) in Section 4. We take p = 4 so that these equations can easily be solved by hand. In the course of the calculations, there are two places where square roots must ....

G. Strang, "Wavelet transforms versus Fourier transforms", Bull. Amer. Math. Soc., Vol. 28, No. 2, April 1993, pp. 288-- 305.

....2.3.5 Wavelet Decomposition The theory of wavelets was developed in the mid eighties and since then it has evolved into a valuable mathematical tool not only for image compression, but also for many other applications in signal processing. A simple definition of wavelets was given by Strang [Strang93]: A function W(x) is a wavelet if the translations and dilations of W(x) are mutually orthogonal. The function W(x) is called mother wavelet. The set of translations and dilations of W(x) forms an orthornormal basis, with good localization properties in both spatial and frequency domains. ....

....was first developed by Morlet, Grossmann, and Meyer; Daubechies and Mallat are two other researchers with major contributions. The research is ongoing, and the number of papers published on wavelets grows at a fast rate. Some survey or tutorial papers on this subject are [Mallat89] Rioul91] [Strang93], Strang94] 2.3.6 Fractal Compression Fractal compression is based on the theory of fractal geometry, developed in the seventies by Mandelbrot [Mandelbrot77] Fractal objects have very complicated structures (in fact, infinitely complicated) which result from the recursive application of ....

Strang G., "Wavelet transforms versus Fourier transforms", Bulletin of the American Mathematical Society, vol. 28, no. 2, pp. 288-305, April 1993.

....density and function estimation, time series, long range dependence, and change point detection. Readers interested in mathematical and practical aspects of wavelets are directed to the monographs of Chui (1992) Daubechies (1992) and Meyer (1992) For an elementary introduction to wavelets see Strang (1993) and Vidakovic and M uller (1999) For the statistical analysis using wavelets see Ogden (1997) and Vidakovic (1999) For reviews of the uses of wavelets in statistics and time series see Morettin (1997) Antoniadis (1997) and Nason and von Sachs (1999) Given a mother wavelet (t) we construct a ....

Strang, G. (1993). Wavelet transforms versus Fourier transforms, Bulletin (New Series) of the American Mathematical Society, 28(2), 288-305.

....to the property that, by choosing the mother wavelet with corresponding regularity properties, one can generate an unconditional wavelet basis in a wide set of function spaces, such as Besov #see Section 2.2 below# or Triebel spaces. For a clear and accessible introduction to wavelets we refer to Strang #1993#. Jawerth Sweldens #1994# provide an excellentoverview of wavelet based multiresolution analyses. Meyer #1992# and Daubechies #1992# give detailed expositions of the mathematical aspects of wavelets. In many practical situations, the functions involved are only de#ned on a compact set, such as ....

Strang, G. #1993#. Wavelet transforms versus Fourier transforms. Bulletin #New Series# of the American Mathematical Society 28, 288-305.

....k Gamma1 X i= Gamma1 1 X j= Gamma1 h ij ; fi ij : 14 Mallat denotes this operator by A d i instead. CHAPTER 3. TRANSFORM METHODS 45 We will not delve into the process of finding this unique scaling function and mother wavelet pair: the reader is referred instead to [16] 29] 5] [25]. However, let us give some typical examples of wavelet scaling function pair. The natural multiresolution extension of the usual pixel representation, is the Haar wavelet decomposition. It is characterized by OE(t) 8 : 1 if t 2 [0; 1[ 0 otherwise (t) 8 : 1 ....

Gilbert Strang. Wavelet transforms versus fourier transforms. Bulletin (New Series) of the American Mathematical Society, 28(2), April 1993. BIBLIOGRAPHY 128

....wavelet frames and orthonormal wavelets. Wavelet frame theory gives an over complete or redundant transform. Orthonormal wavelet theory (also called multiresolution analysis theory) gives an orthonormal, non redundant transform. There are now a number of texts and tutorials available on wavelets [20,31,72,90,91]. For the rest of this chapter, I will concentrate on the discrete orthonormal wavelet transform. 3.1 Discrete Orthonormal Wavelets One Dimension The discrete orthonormal wavelet transform (DWT) is defined as a series of coefficients DWT[m;n] resulting from the projection of a signal f(x) 2 L ....

Gilbert Strang. Wavelet transforms versus Fourier transforms. Bulletin of the American Mathematical Society, 28(2):288--305, 1993.

.... A t ( e.g. the oscillatory and locally stationary processes of Priestley (1981) and Dahlhaus (1997) respectively) Our approach is different in that we replace the set of harmonics fexp(i t)j 2 [ Gamma ; g by a set of discrete non decimated wavelets (see Nason and Silverman (1994) or Strang (1993) for introductions to wavelets) Recently, local atomic decompositions (wavelets, wavelet libraries) have become popular for the analysis of deterministic signals as alternatives to non local Fourier representations (Rioul and Vetterli (1991) Flandrin (1993) The question immediately arises: is ....

Strang, G. (1993) Wavelet transforms versus Fourier transforms. Bull. (New Series) Am.

....may be unable to effectively analyze time series with discontinuities, or whose structure changes with time. In contrast, the wavelet transform maintains information in the transform about both scale and location. For an excellent comparison of wavelet transforms and Fourier transforms see Strang (1993). New work in this chapter include investigating the effect of different boundary conditions on the discrete wavelet transform coefficients, and comparing the phase functions associated with wavelets in current use. 6.2 The wavelet transform We can think of wavelets as forming an orthonormal ....

Strang G. (1993). Wavelet transforms versus Fourier transforms. Bulletin (New Series) of the American Mathematical Society 28, 288--305.

....shrinkage is briefly reviewed in Section 2. Section 3 introduces the cross validation algorithms and Section 4 illustrates the algorithms in one and two dimensions using simulations and by application to some real data collected on breathing patterns. For further information on wavelets see Strang (1993), who provides an accessible introduction, and Nason and Silverman (1994) who discuss wavelets in a statistical context. Meyer (1992) and Daubechies (1992) both give detailed expositions of the mathematical aspects of wavelets. 2 Wavelet Function Estimation 2.1 Wavelets Wavelet estimators may be ....

Strang, G. (1993) Wavelet transforms versus Fourier transforms. Bull. (New Series) Am. Math.

....v, if the matrix (u) is nonsingular. The matrix algebra is important in the analysis of the spectral properties of band Toeplitz matrices and the computation of their tensor rank, as well as in solving Toeplitz and block Toeplitz linear systems (see Section 4. 2) The reader is referred to [Strang, 1993, Bini and Favati, 1993] on wavelet and Hartley transforms. 3 Fast Polynomial and Integer Arithmetic Computations with integers and polynomials, in one or more variables, is of fundamental importance in computational mathematics and computer science. Such operations lie at the core of every ....

Strang, G. 1993. Wavelet transforms versus Fourier transforms. Bulletin (New Series) of the American Mathematical Society, 28(2):288-305.

....to have them as part of their toolbox if an appropriate new area of possible application comes along The aim of this paper is to provide an eclectic personal view, rather than an exhaustive survey, of wavelet methods. For those who would like to read more, a good introduction is provided by Strang (1993). More detailed treatments, in increasing order of mathematical sophistication, are given by Chui (1992) and Daubechies (1992) for example. A key computational aspect of wavelet methods is the discrete wavelet transform, which is based on filtering ideas discussed extensively in the engineering ....

Strang, G. (1993). Wavelet transforms versus Fourier transforms. Bull.

.... A t ( e.g. the oscillatory and locally stationary processes of Priestley (1981) and Dahlhaus (1997) respectively) Our approach is different in that we replace the set of harmonics fexp(i t)j 2 [ Gamma ; g by a set of discrete non decimated wavelets (see Nason and Silverman (1994) or Strang (1993) for introductions to wavelets) Recently, local atomic decompositions (wavelets, wavelet libraries) have become popular for the analysis of deterministic signals as alternatives to non local Fourier representations (Rioul and Vetterli (1991) Flandrin (1993) The question immediately arises: is ....

Strang, G. (1993) Wavelet transforms versus Fourier transforms. Bull. (New Series) Am.

....and Dahlhaus (1997) respectively) Our approach is somewhat different in that we replace the set of harmonics fexp(i t)j 2 [ Gamma ; g by a set of locally supported wavelets. Wavelets are small oscillations that are localized in the time and Fourier domains, see Nason and Silverman (1994) or Strang (1993) for an introduction to wavelets. See Daubechies (1992) Meyer (1992) or Chui (1992) for more authoritative expositions) Recently, local atomic decompositions (wavelets, wavelet libraries) Time (hours) 80 100 120 140 160 180 22 23 00 01 02 03 04 05 06 Figure 1: Heart rate recording of 66 day ....

Strang, G. (1993) Wavelet transforms versus Fourier transforms. Bull. (New Series) Am. Math.

....density and function estimation, time series, long range dependence, and change point detection. Readers interested in mathematical and practical aspects of wavelets are directed to monographs of Chui (1992) Daubechies (1992) and Meyer (1992) For an elementary introduction to wavelets see Strang (1993) and Vidakovic and Muller (1994) Let y be a data vector of dimension (size) N = 2 n . Suppose that the vector y is wavelettransformed to a vector d ; i.e. d = Wy : The transformation is linear and orthogonal and can be described by an orthogonal matrix W of dimension N Theta ....

Strang, G. (1993). Wavelet transforms versus Fourier transforms, Bulletin (New Series) of the American Mathematical Society, 28(2), 288-305.

....results in a lowpass image and three detail images) Proceed to the next scale level Code the filterchoice Quantize and code transform coefficients 5 EXPERIMENTAL RESULTS 7 Let N be the number of datapoints to be transformed. The time complexity of the classical wavelet transform is O(N) [17]) therefore our algorithm as well has complexity O(N ) at which the constant is of course bigger as in the wavelet case and depends on the size of the library, the number of decomposition levels and wether NSMRA, IMRA or INSMRA is used. The computation of the cost functions of the detail spaces ....

G. Strang. Wavelet transforms versus fourier transforms. Bull. Amer. Math. Soc., 28(2):288--305, 1993.

....a particular time period of e#ective length # J 1 #t. Background material on wavelets supporting this qualitative description and supplementing the mathematical development in the next three subsections can be found in, e.g. Mallat (1989) Daubechies (1992) Press et al. 1992) Meyer (1993) Strang (1993), Newland (1993) Nason and Silverman (1994) Percival and Guttorp (1994) Percival (1995) McCoy and Walden (1996) and Lindsay et al. 1996) 2.1 Construction of the DWT The DWT matrix W is constructed based entirely upon a wavelet filter h 1,0 , h 1,L1 1 of even length L 1 (Mallat, ....

Strang, G. (1993), "Wavelet Transforms Versus Fourier Transforms," Bulletin of the American Mathematical Society, 28, 288--305.

.... introductions to spectral analysis can be found in Koopmans (1983) Brillinger (1974) Kay and Marple (1981) Jones (1985) De Marchi and Lo Presti (1993) and Percival (1994) while tutorials on wavelets are given in Strang (1989) Rioul and Vetterli (1991) Section 13.10 of Press et al. 1992) Strang (1993), Meyer (1993) Strichartz (1993) and Tewfik et al. 1993) There are numerous books on spectral analysis, including Jenkins and Watts (1967) Koopmans (1974) Bloomfield (1976) Priestley (1981) Kay (1988) and Percival and Walden (1993) The book by Daubechies (1992) discusses the mathematical ....

Strang, G. (1993) Wavelet Transforms versus Fourier Transforms. Bulletin of the American Mathematical Society , 28, 288--305.

....Dahlhaus [8] respectively) Our approach is somewhat different in that we replace the set of harmonics fexp(i t)j 2 [ Gamma ; g by a set of locally supported wavelets. Wavelets are small oscillations that are localized in the time and Fourier domains, see Nason and Silverman [22] or Strang [29] for an introduction Time (hours) 80 100 120 140 160 180 22 23 00 01 02 03 04 05 06 Figure 1: Heart rate recording of 66 day old baby. Series is sampled at 0.25Hz and is recorded from 21:17:59 to 06:27:18. to wavelets. See Daubechies [10] Meyer [19] or Chui [6] for more authoritative ....

G. Strang. Wavelet transforms versus Fourier transforms. Bull. (New Series) Am. Math. Soc. , 28:288--305, 1993.

.... series analysis and signal processing has increased in recent years due to their potential for solving a number of practical problems; for background on wavelets, see, e.g. Mallat (1989) Strang (1989) Rioul Vetterli (1991) Daubechies (1992) Press et al. 1992) Donoho (1993) Meyer (1993) Strang (1993), Strichartz (1993) and Vaidyanathan (1993) In particular, wavelets can decompose the variance of a physical process across di#erent scales and have been used in this way in a number of scientific and engineering disciplines; see Gamage (1990) Bradshaw Spies (1992) Flandrin (1992) Gao Li ....

....2 with f j # j N and Y # 1 N N X t=1 Y t . The periodogram satisfies a sampling version of Equation (1) namely, 1 N N 2 X j= N 2 1 S Y (f j ) 1 N N X t=1 (Y t Y ) 2 . A fast Fourier transform algorithm can compute S Y using just O(N log 2 (N) multiplications (Strang, 1993), but the periodogram is not a useful estimator of S Y because it can be badly biased and is an inconsistent estimator. To deal with these deficiencies, a practitioner must decide if bias is present and, if so, compensate for it using prewhitening and or tapering, after which the resulting ....

[Article contains additional citation context not shown here]

STRANG, G. (1993). Wavelet transforms versus Fourier transforms. Bull. Amer.

.... traces (see Morlet [16, 17] although the fundamental principle of time and frequency localisation is much older (see Gabor [10] For further general background reading, we cite works by Chui [1, 2] Cohen and Ryan [4] Daubechies [7] Devore and Lucier [8] Meyer [14, 15] and Strang [18, 19]. 2 Multiresolution and scaling functions We work throughout in the context of the space L 2 (R) of real valued functions. Suppose that we have an infinite sequence of nested closed subspaces f0g ae Delta Delta Delta ae V Gamma1 ae V 0 ae V 1 ae V 2 ae Delta Delta Delta ae L 2 (R) 2.1) ....

G. Strang, 1993. Wavelet transforms versus Fourier transforms. Bull. Amer. Math. Soc., 28: 288--305.

.... and Abramovich and Silverman (1997) classification, see Saito (1994) Coifman and Saito (1994) Buckheit and Donoho (1995) Learned and Willsky (1995) and Saito and Coifman (1996) For a general statistical introduction to wavelets see Nason and Silverman (1994) and Bruce and Gao (1996) or see Strang (1993) for a more mathematical view. More detailed comprehensive expositions are Daubechies (1992) Meyer (1992) and Chui (1992) Wavelets are a type of building block for constructing functions. More precisely wavelets form bases for function spaces such as L 2 (R) This article only considers ....

Strang, G. (1993). Wavelet transforms versus Fourier transforms. Bull. (New Series) Am. Math. Soc., 28, 288--305.

....a publicly available package of routines, Nason (1993) for the statistical language S. The package is called wavethresh and is available from the StatLib archive. Appendix A gives full instructions on how to obtain the wavethresh package. A gentle introduction to wavelet methods is provided by Strang (1993). For a more detailed discussion the reader is referred, for example, to Daubechies (1992) and Chui (1992) The statistical aspects of the package are mainly due to Donoho and Johnstone (1993) In this paper we concentrate on the discrete School of Mathematics, University of Bristol, University ....

....jk where Delta; Delta denotes inner product. Clearly, given a function we will wish to compute its wavelet coefficients. The software described later computes wavelet coefficients but uses a discretized version of (5) 2. 1 What can wavelets offer This question is discussed in detail by Strang (1993) so we will say relatively little here. In the Fourier transform of a function f on ( Gamma1; 1) b f ( Z 1 Gamma1 f(t) exp( Gammai t) dt; we usually identify t with time and with frequency. To obtain information about a particular frequency, we have to integrate over the whole domain ....

Strang, G. (1993). Wavelet transforms versus Fourier transforms. Bulletin (New Series) of the American Mathematical Society, 28(2), 288--305.

....to the property that, by choosing the mother wavelet with corresponding regularity properties, one can generate an unconditional wavelet basis in a wide set of function spaces, such as Besov (see Section 2.2 below) or Triebel spaces. For a clear and accessible introduction to wavelets we refer to Strang (1993). Jawerth Sweldens (1994) provide an excellent overview of wavelet based multiresolution analyses. Meyer (1992) and Daubechies (1992) give detailed expositions of the mathematical aspects of wavelets. In many practical situations, the functions involved are only defined on a compact set, such as ....

Strang, G. (1993). Wavelet transforms versus Fourier transforms. Bulletin (New Series) of the American Mathematical Society 28, 288-305.

....and Dahlhaus (1997) respectively) Our approach is somewhat different in that we replace the set of harmonics fexp(i t)j 2 [ Gamma ; g by a set of locally supported wavelets. Wavelets are small oscillations that are localized in the time and Fourier domains, see Nason and Silverman (1994) or Strang (1993) for introductions. See Daubechies (1992) Meyer (1992) or Chui (1992) for more comprehensive expositions) Recently, local atomic decompositions (wavelets, wavelet libraries) have become popular for the analysis of deterministic signals as alternatives to non local Fourier representations (Rioul ....

Strang, G. (1993) Wavelet transforms versus Fourier transforms. Bull. (New Series) Am. Math.

....can be found in [75, 76, 77, 31, 32, 73] Walter [79] gives a good introduction to wavelets in the context of other orthogonal systems. Strang [66] gives a readable, non technical description of wavelets and applications. For a comparison of wavelet transforms and Fourier transforms, see Strang [67]. For a discussion of time frequency localization in signal processing, see [15] 8 1.4 Motivation This dissertation is motivated mainly by an ever present need for more efficient techniques for analyzing, manipulating and understanding the visual world around us using computers. The problems ....

....as f(x) 1 C Z 1 Gamma1 Z 1 Gamma1 W(a; b) a;b (x) dadb a 2 : 2. 17) From the admissibility condition, it follows that (0) 0 and we see why the name wavelet is appropriate for describing (x) A more complete discussion of wavelet transforms and their properties may be found in [23, 30, 58, 67, 80, 81]. 2.3.3 The Dyadic Discrete Wavelet Transform In the continuous wavelet transform (Eq. 2.13) if a is restricted to powers of 2, and b to integer multiples of a, we have the dyadic discrete wavelet transform (DWT) Wf(j; k) 2 Gammaj =2 Z 1 Gamma1 f(x) 2 Gammaj x Gamma k) j; k 2 Z : ....

G. Strang. Wavelet Transforms versus Fourier Transforms. Bull. Am. Math. Society. 28, pp. 288--305, 1993.

....to be already somewhat familiar with this material, and we include it primarily to establish the notation which we will use for the remainder of the paper. For thorough introductions to subband and wavelet coding, the reader is referred to a plethora of books, tutorials, and survey articles [157, 149, 127, 141, 47, 83]. Section 2 also discusses some issues of entropy coding, scalar quantization, and bit allocation as they apply to subband image coding systems. Section 3 presents a brief overview of several VQ techniques that have been studied in the context of subband coding. Section 4 presents a survey of a ....

Gilbert Strang. Wavelet transforms versus fourier transforms. Bulletin (New Series) of the AMS, 28(2):288--305, April 93.

....by Nason and Silverman [14] Each of these reconstructions was obtained by applying a wavelet transform to the data, applying the universal filter of Donoho and Johnstone [6] and then applying an inverse wavelet transform. In Subplot E, Haar wavelets are used in the transformations (see [16] or [13] for a discussion of Haar wavelets) These wavelets are generated by a discontinuous mother wavelet and are of regularity level 1 (see [14] This reconstruction clearly maintains the discontinuities of the true image; however, there appear to be extraneous effects similar to ringing which ....

G. Strang, Wavelet Transforms Versus Fourier Transforms, American Mathematical Society Bulletin, vol. 28 (1993), pp. 288-305.

....shrinkage is briefly reviewed in Section 2. Section 3 introduces the cross validation algorithms and Section 4 illustrates the algorithms in one and two dimensions using simulations and by application to some real data collected on breathing patterns. For further information on wavelets see Strang (1993), who provides an accessible introduction, and Nason and Silverman (1994) who discuss wavelets in a statistical context. Meyer (1992) and Daubechies (1992) both give detailed expositions of the mathematical aspects of wavelets. 2 Wavelet Function Estimation 2.1 Wavelets Wavelet estimators may be ....

Strang, G. (1993) Wavelet transforms versus Fourier transforms. Bull. (New Series) Am. Math.

....Figure 1. Walsh functions. They are often described as discrete analogues of the sine and cosine functions, and, as such, have found applications in electrical engineering for digital signal processing [20, 7] Recent interest in wavelets has focused attention on the closely related Haar functions [25, 22]. There are different ways to describe the Walsh functions and there are alternative indexing conventions. For the purposes of this paper the most convenient construction of the Walsh functions is obtained from the increasing sequence of subspaces Hm ae Hm 1 of L 2 [0; 1] m 2 N, where Hm 80 ....

G. Strang, Wavelet transforms versus Fourier transforms, Bull. Amer. Math. Soc. 28 (1993), 288--305.

....f k 2 2;n ; which we would like to make as small as possible. The remainder of this paper is about building f using wavelets. The next section gives a brief overview of wavelets and their role in regression. 2. 1 Wavelets A general introduction to the theory of wavelets can be found in Strang (1993). An introduction that considers the statistical uses of wavelets can be found in Nason and Silverman (1994) More technical and complete surveys include Daubechies (1992) and Chui (1992) There are many families of wavelets. Each family can be derived from a single special function known as the ....

Strang, G. 1993. Wavelet transforms versus Fourier transforms. Bulletin (New Series) of the American Mathematical Society, 28, 288--305.

....yields a particularly attractive orthonormal basis for R n . The expansion of a vector in terms of this discrete basis is achieved by a very fast algorithm the discrete wavelet transform operates in O(n) steps while the discrete Fourier transform requires O(n log n) The expository paper [S] describes analogies and competitions between these transforms. In most of signal processing the standard methods are Fourier based. In the compression of fingerprint images (the FBI has 25 million to digitize and compare) wavelet bases now seem to be superior. 54 CHRISTOPHER HEIL AND GILBERT ....

G. Strang, Wavelet transforms versus Fourier transforms, Bull. Amer. Math. Soc. 28 (1993), 288--305.

No context found.

Strang, G. #1993#. Wavelet transforms versus Fourier transforms. Bulletin #New Series# of the American Mathematical Society 28, 288-305.

No context found.

Gilbert Strang, Wavelet transforms versus fourier transforms, Bulletin of the AMS 28 (1993), no. 2, 288-305.

No context found.

Strang, G. "Wavelet Transforms Versus Fourier Transforms.", Bull. Amer. Math. Soc. 28, 1993, pp. 288-305

No context found.

Strang G.: Wavelet Transforms versus Fourier Transforms. Bulletin of the American Mathematical Society 28(2), 1993, 288--305.

No context found.

G. Strang, Wavelet transforms versus Fourier transforms, Bull. Amer. Math. Soc. (N.S.), 28 (1993), pp. 288--305.

No context found.

Strang, G. (1993). Wavelet transforms versus Fourier transforms. Bull. (New Series) Am. Math. Soc., 28, 288-305.

No context found.

G. Strang. Wavelet transforms versus Fourier transforms. Bull. Amer. Math. Soc. (N.S.), 28(2):288--305, 1993.

No context found.

Strang, G. (1993). Wavelet transforms versus Fourier transforms. Bull. (New Series) Am. Math. Soc., 28, 288--305. REFERENCES 32

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC